Re: CVS commit: src/sys/dev

[Matthias Scheler <tron%NetBSD.org@localhost>]

On Mon, Nov 09, 2009 at 07:04:41PM -0600, David Young wrote:
> > Log Message:
> > Don't allocate block buffers on the stack. This can cause stack overflows
> > in the kernel and breaks SSP builds.
> 
> This is a step in the right direction, but now cgd(4) is doing
> malloc(..., M_WAITOK) in (software) interrupt context.  A LOCKDEBUG
> kernel will panic.  How can we avoid the stack overflows and such in a
> different way?

1.) I misunderstood the meaning of blocksize in this context. It is
    the not the disk blocksize but the blocksize of the cipher.
    The maximum value at the moment is 128 bytes which will result
    in a total of 384 bytes of buffers allocated on the stack.
    There is therefore no risk of stack overflows.

2.) I have backed out my changes to fix the LOCKDEBUG kernel panics.

Thanks a lot for pointing out the problem.

        Kind regards

-- 
Matthias Scheler                                  http://zhadum.org.uk/